How websites could make more money selling display ads

Professor Carl Mela found an algorithmic way to optimize prices in real-time bids and increase publishers' revenue

Published: Nov 4, 2024 11:23:03 AM IST

Updated: Nov 4, 2024 11:38:08 AM IST

Real-time bidding becoming a major source of revenue for online publishers. Setting the right “reserve price” on real-time auctions may help publishers increase their revenue by more than 30%. Image: Shutterstock

Whenever you load a website on your browser, the ads on that site are selected in a real-time auction that declares a winner within milliseconds.

Such instantaneous auctions — known as “real-time bidding” (RTB) — are projected to produce more than $16B of revenue in 2024, a growing slice of the digital advertising market.

“Ad auctions are the lion’s share of revenue at Meta and Google,” said Carl Mela, the T. Austin Finch Foundation Professor at Duke University’s Fuqua School of Business. “And increasingly, Amazon’s too. Some analysts estimate nearly half of their profitability comes from advertising.”

Despite real-time bidding becoming a major source of revenue for online publishers — including media publishers, such as The New York Times — little research has focused on how to optimize the publisher’s revenues, Mela said.

In a new paper, “Optimizing Reserve Prices in Display Advertising Auctions,” Mela and co-author Hana Choi of the University of Rochester (a Fuqua Ph.D. graduate) show that setting the right “reserve price” on real-time auctions may help publishers increase their revenue by more than 30%.

Asking the right “reserve price”

In an auction, the auctioneer usually sets a reserve price, that is the minimum price they are willing to accept. If the highest bid falls below the reserve, the auctioneer doesn't sell the item.

Setting the right reserve price can make a substantial difference in terms of publisher’s revenue, Mela said, especially considering that display ad auctions often use the “second-highest price” rule — whereby the highest bidder wins the auction but pays the price committed by the second-highest bid.

“Suppose the advertiser is willing to pay $1 and the second highest bid is 50 cents,” Mela said. “The winner is the $1 bid, and they will pay 50 cents. But if the auctioneer sets a reserve price at 75 cents, the advertiser would pay 75 cents, not 50 cents, so the publisher makes more money.”

The key is to set a reserve price that is neither so high that it discourages any bid, nor so low that the publisher leaves a lot of money on the table.

“If the publisher sets its reserve price too high, no bids will be accepted and it won’t sell the ad. If the reserve is set too low, then the publisher could have raised it and increased payments made by the advertiser,” Mela said.

Also read: The CMO Survey: Marketers spend on new technologies as they battle usage and impact challenges

A model to calculate the optimal price for real-time ad bids

In their paper, the researchers developed an algorithm to maximize publishers’ revenue by optimizing the reserve price, based on how much advertisers value the ads shown on the publisher’s site.

“Some companies want their brand to be known, so they are going to spend a little bit more on an ad, even to the point they might lose some revenue in the short run,” Mela said.

The researchers estimated the advertiser value by looking at how much the advertisers had paid for each ad type in the past. Through a dataset of almost 9 million bid and payment observations provided by an ad exchange (a platform matching websites and advertisers, executing real-time bids), the researchers found empirical evidence of how branding goals and other considerations affect the advertisers’ willingness to pay for an ad.

By feeding these estimates into their algorithm, the researchers determined that setting an optimal reserve price would lead to a 35% increase in revenue for publishers in ad exchange auctions.

To validate their predictions, they applied their model to 2019 data from a “large, premium publisher, ranked within the U.S. top 10 by Comscore.” The results of this field experiment confirmed the model’s forecasts.

“Past research had not focused on the publisher pricing problem, perhaps because data on advertiser bidding and publisher reserve prices are difficult to obtain,” Mela said.

Plugging in an algorithm to compute the optimal price

The real-time bidding market for display advertising provides an enormous supply of advertising opportunities across publishers, and a similarly large demand for those advertising slots from advertisers. As a result, ad exchanges instantly match advertisers seeking specific consumers with websites offering the right customer profile.

“The moment a consumer visits a website page — say, the Wall Street Journal — an advertising opportunity is created,” Mela said. ”At that moment, there's a number of advertisers who want to buy that advertising opportunity. An auction is conducted in which multiple advertisers bid for that ad slot, the winner is declared, and a payment is sent from the advertiser to the publisher. And the ad copy is sent from the advertiser into that display ad on the webpage. Remarkably, this entire process typically occurs within 100 milliseconds.”

The programmatic nature of this “continuous ad-buying game” gives publishers a unique opportunity to employ algorithms to enhance revenues, without having to invest in new infrastructure, Mela said.

All publishers need to do is employ an algorithm to set the optimal price, he said, an operation most companies are already able to perform.

“In many other business contexts, increasing revenue requires capital investment or more employees. In real-time bidding, companies can use our algorithm to set the optimal price with minimal programming costs. Most tech companies have people with the skills to do that,” Mela said.

[This article has been reproduced with permission from Duke University's Fuqua School of Business. This piece originally appeared on Duke Fuqua Insights]